On Wed, 9 Sep 2020 at 13:37, Wolfram Sang <wsa+renesas@xxxxxxxxxxxxxxxxxxxx> wrote: > > Hi Ulf, > > > > Hmm, there are some wireless drivers using it as well. I am confused, is > > > this considered "upper layer"? > > > > > > drivers/net/wireless/ath/ath10k/sdio.c: ret = mmc_hw_reset(ar_sdio->func->card->host); > > > drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.c: mmc_hw_reset(sdiodev->func1->card->host); > > > drivers/net/wireless/marvell/mwifiex/sdio.c: ret = mmc_hw_reset(func->card->host); > > > drivers/net/wireless/ti/wlcore/sdio.c: mmc_hw_reset(card->host); > > > > Correct, these are "upper layers". The same applies for the mmc block > > device driver. > > > > In this way there is a guarantee that the struct mmc_card is still present. > > Ah, now I get it. "upper layers" as in consumers. And because consumers > sit on a card, this guarantees that mmc_card is still there. Correct? Yes. > > > That would be great. I appreciate all kinds of improvements on the doc parts. > > You are welcome! > > > Perhaps a better option is to return a specific error code for the > > last request, that makes the core run mmc_hw_reset(). Or potentially, > > add a host cap and let the core treat some error code, specifically > > for hosts like tmio. > > A specific errno could work. I don't see the advantage of a CAP (besides > it is rather a quirk than a cap). We could also have > 'mmc_controller_card_reset()' or something which ensures mmc_card is > present and let that controllers call when they see fit. Or? Maybe something like "mmc_controller_card_reset" could work, but it's not going to be that straight forward. In the end, we depend on the context for when the host driver would call such a function. In some cases it must call mmc_claim_host() while in others it shouldn't. BTW, I see that tmio_mmc_reset() is called at tmio_mmc_host_runtime_resume(). This seems to work fine without having to make a full reset of the card. Why can't you do something similar to that instead? Kind regards Uffe
